The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted audio and video data from various sources, including file uploads and remote URLs. This input serves as a potential vector for indirect prompt injection, where malicious instructions spoken in the media could affect the agent's subsequent actions.
Ingestion points: Transcription methods accepting file, source_url, cloud_storage_url, and the audio_base_64 parameter in real-time streaming (SKILL.md, references/transcription-options.md).
Boundary markers: The provided code examples do not demonstrate the use of delimiters or specific instructions to the agent to ignore commands within the transcribed text.
Capability inventory: While the skill primarily outputs text, the agent consuming this text may have extensive file system, network, or tool access.
Sanitization: No sanitization or filtering of transcribed text is implemented in the reference code.
[REMOTE_CODE_EXECUTION]: The documentation includes a command to install the vendor's CLI via a shell script downloaded from GitHub and piped directly to the shell.
Evidence: curl --proto '=https' --tlsv1.2 -LsSf https://github.com/elevenlabs/cli/releases/latest/download/elevenlabs-cli-installer.sh | sh (references/installation.md).
Context: The resource is hosted on the official GitHub repository of the skill's author, which is a standard distribution method for this vendor.
[EXTERNAL_DOWNLOADS]: The skill documentation guides users to install various software packages and utilities from external registries and repositories.
Evidence: References to official NPM packages (@elevenlabs/elevenlabs-js), PyPI packages (elevenlabs), and platform-specific package managers like Homebrew and Scoop (references/installation.md).
Context: All external dependencies are official vendor-maintained resources.

speech-to-text